sebastianspicker.com

IT engineer, researcher, and machine learning specialist.

Say hi: hi@sebastianspicker.com

Identity

Sebastian J. Spicker

IT Engineer · Researcher · Machine Learning

IT engineer at the Cologne University of Music and Dance, working at the intersection of IT, music technology, physics, pedagogics, and AI. Research covers experimental astronomy education — smartphone observatories, remote laboratory design, inquiry-based learning — and the structural analysis of language model failure modes.

Scientific consultant in photovoltaics and energy storage. The approach to inclusive and heterogeneous learning environments is practical rather than rhetorical, informed by direct experience as someone on the autism spectrum. Outside work, member of the Cologne Cat Protection Society.

Snapshot

Short CV

2020–present
IT Engineer
Cologne University of Music and Dance
2024–2025
Scientific Consultant
CMD Global
2016–2020
Research Associate, Physics Education
University of Cologne

Focus

Areas

  • Astronomy and physics education
  • Experimental and remote laboratory design
  • Learning design and curriculum development
  • Public astronomy outreach and science communication
  • Heterogeneity and inclusion in education
  • Digital music education and artistic practice
  • Machine learning and language model research
  • Photovoltaics and energy storage

Methods

How I work

  1. Any problem — pedagogical, technical, or structural — gets decomposed into its essential variables and constraints before a solution is attempted.
  2. Physics, music theory, and computer science share formal structures that rarely appear in the same curriculum. Working across all three routinely surfaces analogies and solution paths that discipline-internal thinking misses.
  3. Prototypes are treated as experiments: a lesson format, a lab setup, or a software sketch is a testable hypothesis, not a draft.
  4. If something cannot be explained clearly to a student, the understanding itself is incomplete. Teaching is part of the research process, not a downstream step.
Tools
PythonMachine learning & AILearning techMusic techSensors

Navigation

Constellation Map

Index

Selected Work

Smartphone Planetarium Design

Modular astronomy instrument built from smartphone sensors. Students assemble, calibrate, and collect data from real celestial observations. Part of astro-lab.app — a running research collaboration across three universities. The student is simultaneously experimenter and instrument maker.

The instrument uses the smartphone camera for exoplanet transit photometry and onboard sensors for orientation and alignment. The measurement is real — not a simulation.

Inquiry-based design: students formulate the observation hypothesis, handle calibration, and identify sources of error independently. Data contributes to a shared cross-institutional dataset.

Low-cost hardware by design. Removing the mystification of scientific instruments is part of the pedagogy — if students can build it themselves, they understand what it measures.

Language Model Failure Modes

Original research into the structural limits of large language models: vocabulary incompleteness, grounding failures, and pragmatic inference failures. Each is a formally distinct failure type — not malfunction, but architectural constraint.

Vocabulary incompleteness: models conflate world knowledge with output vocabulary. Logit lens analysis shows the correct semantic concept is represented internally — but the corresponding token does not exist in the fixed vocabulary, so the model approximates without flagging the gap.

Grounding failure: models cannot access current world-state. Asked whether to drive to a car wash in rain, the model averages over training contexts — not a knowledge gap, but an architecturally absent access to 'right now'.

Pragmatic inference failure: the model has all required information but applies literal semantics where cooperative intent inference (Grice's maxims) is required. Each failure type has a different formal cause and requires a different mitigation.

Reflection-based Artistic Practice Lab

Research project applying scientific method to music conservatory pedagogy. Developed structured reflection formats for artistic professional practice — importing empirical rigor into a domain where artistic intuition is the primary instrument.

Music students applied structured reflection protocols to their own artistic decision-making — treating practice sessions as data. The underlying question: can making implicit processes explicit improve artistic development?

The connection to physics pedagogy is direct: both domains combine formal structure (voice leading rules ≈ physical laws) with exploratory practice. Research methods transfer in both directions.

Produced reflection formats now integrated into conservatory pedagogy at HfMT Cologne. Cologne University of Music and Dance, 2020–2023.

Inclusive Remote Physics Modules★ Highlight

Remote physics lessons designed with consistent structure, minimal cognitive overhead, and predictable pacing. Tested with mixed-ability secondary cohorts. The design draws on direct experience of how structural and environmental factors affect learning.

Designed from scratch rather than adapted from standard lessons — adaptation tends to preserve the hidden assumptions that create barriers in the first place.

Structural principles: fixed opening routine, consistent visual layout, explicit advance notice of all transitions. Pacing aligned to processing speed rather than curriculum coverage.

Iterated from direct observation: moments where students disengaged were treated as design failures, not student failures.

Archive

Publications & Talks

2024
2024
Sensorbasierte Smartphone-Experimente rund um die Exoplaneten-Lichtkurve.
INVITED TALK · Astronomische Gesellschaft annual meeting
talk
Exoplanet Hunting in the Classroom: Video-aided light curve analysis with smartphones.
PAPER · The Physics Teacher 62(3), 226–229
journal
2023
2023
Wie schnell ist schnell genug? Digitale Arbeit mit LoLA und MVTP.
PAPER · Cologne University of Music and Dance
report
Astronomische Smartphone-Experimente rund um die Exoplaneten-Lichtkurve.
INVITED TALK · National astronomy teacher training
talk
Manual zur Erstellung von Unterrichtsvideographien.
MANUAL · University of Cologne
report
Analogieexperiment zur Suche nach Exomonden.
PAPER · MNU-Journal 76(3), 223–227
journal
2022
2022
astro-lab@home — Bringing science to the sofa.
PAPER · CAP Journal 31, 37–41
journal

Timeline

Projects

2020–present
2020–present
astro-lab.app
Institute of Physics Education, University of Cologne, RWTH Aachen University & University Paris-Saclay.
2020–2023
2020–2023
Reflection based Artistic Professional Practice (RAPPLab)
Cologne University of Music and Dance.
2020–2022
2020–2022
digital student labs: astro-lab@home, astro-lab@school
Institute of Physics Education, University of Cologne.
2016–2020
2016–2020
Future Strategy for Teacher Education Cologne
Shaping Heterogeneity and Inclusion, CompetenceLabs: MediaLab, University of Cologne.
2016–2019
2016–2019
zdi student lab “Astronomy”
Institute of Physics Education, University of Cologne.

Contact

Home base

Uplink: hi@sebastianspicker.com

Ground station: Cologne, Germany

Coordinates: J2000.0 · 50.9375°N, 6.9603°E